Is This Image AI? How to Check Content Credentials (C2PA)

Content Credentials are a signed 'digital nutrition label' attached to an image. They are built on the C2PA standard (the Coalition for Content Provenance and Authenticity), and they record who or what made the file, including whether AI was involved. As of 2026, most of the big AI generators add them automatically: DALL·E (ChatGPT), Adobe Firefly, Microsoft Designer, and Google Gemini all sign their output. So the fastest way to answer 'is this AI?' is often just to read the credential the generator already attached.

Here are the three ways to check, from easiest to most thorough, and then the honest limitations, because Content Credentials answer the question only when they survive, and they often do not.

Method 1: Look for the 'CR' icon

On platforms that support Content Credentials, a small 'CR' badge appears in the corner of the image. LinkedIn shows it on content that carries credentials; so do a growing number of news sites and creative tools. Click or tap the icon and it expands to show the source, the tool used, and whether AI was involved. If you see a CR badge, you can read the provenance without any extra tool. The catch: most places, including the majority of social feeds, do not display the icon even when the credential is present, so its absence proves nothing.

Method 2: Inspect the manifest with a viewer (the reliable one)

The dependable method is to read the file's C2PA manifest directly. Drop the image into a Content Credentials viewer and it tells you, in plain terms: is a signed credential present, who the claim generator was (for example 'DALL·E 3' or 'Adobe Firefly'), and whether the manifest carries an AI-generation assertion. A standardized AI marker like 'trainedAlgorithmicMedia' means fully AI-generated; 'compositeWithTrainedAlgorithmicMedia' means AI was used to edit or composite. This is the method that gives you a definite answer when the credential is intact.

Use the C2PA viewer for this. It reads the manifest in your browser, shows the claim generator and the AI flag, and never uploads the file. If it reports a signed manifest with an AI assertion from a known generator, that is strong evidence the image is AI-made.

Method 3: When there is no credential, fall back to your eyes (and other tools)

If the viewer finds no credential, you cannot conclude anything from C2PA alone, because plenty of AI images never had one or had it stripped (more on that below). At that point you are back to the older signals: look for the classic AI artifacts (garbled text, extra or merged fingers, melted backgrounds, impossible reflections, overly smooth skin), run a reverse-image search to find the origin, and check ordinary EXIF metadata for a camera make and model (a real photo usually has camera EXIF; a pure AI image usually does not). For Google-generated media in 2026, a SynthID watermark may also be present in the pixels even when metadata is gone, though reading SynthID needs Google's own detector.

The big catch: screenshots and uploads erase the evidence

This is the part most people miss. A Content Credential lives in the file's metadata, so anything that creates a new file or re-encodes the image can remove it:

• A screenshot destroys it completely. Screenshotting an AI image produces a brand-new file with no manifest, so the provenance chain is gone. If someone sends you a screenshot, C2PA can tell you nothing.
• Most social uploads strip it. Many platforms re-encode images on upload and drop the C2PA manifest in the process. Some, like LinkedIn and TikTok, preserve and even display it, but most do not, so an image saved from a feed often arrives with no credential.
• Editing in non-C2PA software removes it. Re-saving through a tool that does not write Content Credentials wipes the manifest.
• Not every AI tool adds one. Midjourney (as of 2026) and most local Stable Diffusion setups do not embed C2PA at all, so their output never had a credential to find.

So the honest summary: a present, valid Content Credential is strong proof, but an absent one is not proof of anything. 'No credential' can mean human-made, or AI-made-and-stripped, or AI-made-by-a-tool-that-does-not-sign. Use C2PA as your first and strongest check, then fall back to artifacts, reverse search, and EXIF when it comes up empty.

Want to remove credentials instead of read them?

The flip side: if you want to strip Content Credentials and other metadata from your own images, for privacy, before sharing, the metadata cleanup tool removes the C2PA manifest along with EXIF, GPS, and XMP. One honest caveat, the same one that applies to detection: it removes metadata only. A pixel-level watermark such as Google SynthID is baked into the image data, not the metadata, so no metadata tool can touch it.

The short version

To check if an image is AI in 2026: first read its Content Credential. Look for a CR icon, or drop the file into a C2PA viewer to see the signed manifest, the claim generator, and the AI flag. A valid credential from a known generator is strong proof it is AI. But a screenshot or a social upload usually erases the credential, and some AI tools never add one, so an absent credential proves nothing, fall back to AI artifacts, reverse-image search, and EXIF.

Start with the C2PA viewer to read an image's credentials, and the EXIF and metadata remover if you want to strip your own. For the deeper background, see what Content Credentials are and why they matter.